Theories of intergroup relations are often multilevel theories. Social dominance theory explicitly describes multiple levels of analysis. It predicts that ideologies predict behavior at the individual level, which then contribute to intergroup inequality at the societal level. Social identity theory also outlines multiple levels of analysis. It argues that people's social identification is a function of structural relations between groups (status hierarchies, group permeability). Social role theory also highlights the importance of multiple levels of analysis. The gendered division of labor (i.e., men and women's disproportationate allocation to social roles) predicts the content of gender stereotypes. Thus, positional inequality at the societal level predicts stereotype knowledge at the individual level.
Many of these theories major assumptions are multilevel in nature, highlighting the importance of inequality (in status, power, and position) in predicting individual level prejudice and stereotypes. Thus, multilevel modeling is necessary to test these assumptions empirically.
Multilevel modeling allows us to test various theories of intergroup relations. Variables can vary at different levels of analysis (e.g., the individual and the national levels). For example, individuals can differ in how sexist they are. Some people are sexist, and other people are not sexist. In addition to individual level variability, nations can differ in how sexist they are on average. Some nations, like many Scandinavian nations, are not sexist, and other nations, such as India or Mali, are sexist on average. If we collect cross-national data measuring people's sexist beliefs and policy attitudes, we can calculate correlation matrices for the individual level and the national level. Using the variability at two levels of analysis allows us to construct multilevel models, and we can answer a variety of questions. For example, we can ask whether the strength of the relationship between sexist beliefs and support for women's rights at the individual level changes based on how normative sexism is at the national level. This question could be answered with a cross-level interaction between the slope at the individual level and the norm variable at the national level. In this manner, multilevel modeling allows us to answer theoretical questions in a way that was not possible with other statistical techniques.